4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2017, Intel Corporation.
26 * This file is part of Lustre, http://www.lustre.org/
28 * Implementation of cl_device, cl_req for MDC layer.
30 * Author: Mikhail Pershin <mike.pershin@intel.com>
33 #define DEBUG_SUBSYSTEM S_MDC
35 #include <obd_class.h>
36 #include <lustre_osc.h>
37 #include <uapi/linux/lustre/lustre_param.h>
39 #include "mdc_internal.h"
41 static void mdc_lock_build_policy(const struct lu_env *env,
42 union ldlm_policy_data *policy)
44 memset(policy, 0, sizeof *policy);
45 policy->l_inodebits.bits = MDS_INODELOCK_DOM;
48 int mdc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data)
50 return osc_ldlm_glimpse_ast(dlmlock, data);
53 static void mdc_lock_build_einfo(const struct lu_env *env,
54 const struct cl_lock *lock,
55 struct osc_object *osc,
56 struct ldlm_enqueue_info *einfo)
58 einfo->ei_type = LDLM_IBITS;
59 einfo->ei_mode = osc_cl_lock2ldlm(lock->cll_descr.cld_mode);
60 einfo->ei_cb_bl = mdc_ldlm_blocking_ast;
61 einfo->ei_cb_cp = ldlm_completion_ast;
62 einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
63 einfo->ei_cbdata = osc; /* value to be put into ->l_ast_data */
66 static void mdc_lock_lvb_update(const struct lu_env *env,
67 struct osc_object *osc,
68 struct ldlm_lock *dlmlock,
71 static int mdc_set_dom_lock_data(struct ldlm_lock *lock, void *data)
75 LASSERT(lock != NULL);
76 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
78 lock_res_and_lock(lock);
80 if (lock->l_ast_data == NULL)
81 lock->l_ast_data = data;
82 if (lock->l_ast_data == data)
85 unlock_res_and_lock(lock);
90 int mdc_dom_lock_match(const struct lu_env *env, struct obd_export *exp,
91 struct ldlm_res_id *res_id, enum ldlm_type type,
92 union ldlm_policy_data *policy, enum ldlm_mode mode,
93 __u64 *flags, struct osc_object *obj,
94 struct lustre_handle *lockh,
95 enum ldlm_match_flags match_flags)
97 struct obd_device *obd = exp->exp_obd;
98 __u64 lflags = *flags;
103 rc = ldlm_lock_match_with_skip(obd->obd_namespace, lflags, 0,
104 res_id, type, policy, mode, lockh, match_flags);
106 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
110 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
112 LASSERT(lock != NULL);
113 if (mdc_set_dom_lock_data(lock, obj)) {
114 lock_res_and_lock(lock);
115 if (!ldlm_is_lvb_cached(lock)) {
116 LASSERT(lock->l_ast_data == obj);
117 mdc_lock_lvb_update(env, obj, lock, NULL);
118 ldlm_set_lvb_cached(lock);
120 unlock_res_and_lock(lock);
122 ldlm_lock_decref(lockh, rc);
131 * Finds an existing lock covering a page with given index.
132 * Copy of osc_obj_dlmlock_at_pgoff() but for DoM IBITS lock.
134 struct ldlm_lock *mdc_dlmlock_at_pgoff(const struct lu_env *env,
135 struct osc_object *obj, pgoff_t index,
136 enum osc_dap_flags dap_flags)
138 struct osc_thread_info *info = osc_env_info(env);
139 struct ldlm_res_id *resname = &info->oti_resname;
140 union ldlm_policy_data *policy = &info->oti_policy;
141 struct lustre_handle lockh;
142 struct ldlm_lock *lock = NULL;
145 enum ldlm_match_flags match_flags = 0;
149 fid_build_reg_res_name(lu_object_fid(osc2lu(obj)), resname);
150 mdc_lock_build_policy(env, policy);
152 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
153 if (dap_flags & OSC_DAP_FL_TEST_LOCK)
154 flags |= LDLM_FL_TEST_LOCK;
156 if (dap_flags & OSC_DAP_FL_AST)
157 match_flags |= LDLM_MATCH_AST;
159 if (dap_flags & OSC_DAP_FL_CANCELING)
160 match_flags |= LDLM_MATCH_UNREF;
163 /* Next, search for already existing extent locks that will cover us */
164 /* If we're trying to read, we also search for an existing PW lock. The
165 * VFS and page cache already protect us locally, so lots of readers/
166 * writers can share a single PW lock. */
167 mode = mdc_dom_lock_match(env, osc_export(obj), resname, LDLM_IBITS,
168 policy, LCK_PR | LCK_PW | LCK_GROUP, &flags,
169 obj, &lockh, match_flags);
171 lock = ldlm_handle2lock(&lockh);
172 /* RACE: the lock is cancelled so let's try again */
173 if (unlikely(lock == NULL))
181 * Check if page @page is covered by an extra lock or discard it.
183 static bool mdc_check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
184 struct osc_page *ops, void *cbdata)
186 struct osc_thread_info *info = osc_env_info(env);
187 struct osc_object *osc = cbdata;
190 index = osc_index(ops);
191 if (index >= info->oti_fn_index) {
192 struct ldlm_lock *tmp;
193 struct cl_page *page = ops->ops_cl.cpl_page;
195 /* refresh non-overlapped index */
196 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
197 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
199 info->oti_fn_index = CL_PAGE_EOF;
201 } else if (cl_page_own(env, io, page) == 0) {
202 /* discard the page */
203 cl_page_discard(env, io, page);
204 cl_page_disown(env, io, page);
206 LASSERT(page->cp_state == CPS_FREEING);
210 info->oti_next_index = index + 1;
215 * Discard pages protected by the given lock. This function traverses radix
216 * tree to find all covering pages and discard them. If a page is being covered
217 * by other locks, it should remain in cache.
219 * If error happens on any step, the process continues anyway (the reasoning
220 * behind this being that lock cancellation cannot be delayed indefinitely).
222 static int mdc_lock_discard_pages(const struct lu_env *env,
223 struct osc_object *osc,
224 pgoff_t start, pgoff_t end,
227 struct osc_thread_info *info = osc_env_info(env);
228 struct cl_io *io = &info->oti_io;
229 osc_page_gang_cbt cb;
234 io->ci_obj = cl_object_top(osc2cl(osc));
235 io->ci_ignore_layout = 1;
236 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
240 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
241 info->oti_fn_index = info->oti_next_index = start;
243 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
244 end, cb, (void *)osc);
250 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
251 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
259 if (mode == CLM_WRITE) {
260 result = osc_cache_writeback_range(env, obj, start, end, 1,
262 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
263 obj, start, end, result,
264 discard ? "discarded" : "written back");
269 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
270 rc = mdc_lock_discard_pages(env, obj, start, end,
271 mode == CLM_WRITE || discard);
272 if (result == 0 && rc < 0)
278 void mdc_lock_lockless_cancel(const struct lu_env *env,
279 const struct cl_lock_slice *slice)
281 struct osc_lock *ols = cl2osc_lock(slice);
282 struct osc_object *osc = cl2osc(slice->cls_obj);
283 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
286 LASSERT(ols->ols_dlmlock == NULL);
287 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
290 CERROR("Pages for lockless lock %p were not purged(%d)\n",
293 osc_lock_wake_waiters(env, osc, ols);
297 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
298 * and ldlm_lock caches.
300 static int mdc_dlm_blocking_ast0(const struct lu_env *env,
301 struct ldlm_lock *dlmlock,
304 struct cl_object *obj = NULL;
307 enum cl_lock_mode mode = CLM_READ;
311 LASSERT(flag == LDLM_CB_CANCELING);
312 LASSERT(dlmlock != NULL);
314 lock_res_and_lock(dlmlock);
315 if (dlmlock->l_granted_mode != dlmlock->l_req_mode) {
316 dlmlock->l_ast_data = NULL;
317 unlock_res_and_lock(dlmlock);
321 discard = ldlm_is_discard_data(dlmlock);
322 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
325 if (dlmlock->l_ast_data != NULL) {
326 obj = osc2cl(dlmlock->l_ast_data);
329 unlock_res_and_lock(dlmlock);
331 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
332 * the object has been destroyed. */
334 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
336 /* Destroy pages covered by the extent of the DLM lock */
337 result = mdc_lock_flush(env, cl2osc(obj), cl_index(obj, 0),
338 CL_PAGE_EOF, mode, discard);
339 /* Losing a lock, set KMS to 0.
340 * NB: assumed that DOM lock covers whole data on MDT.
342 /* losing a lock, update kms */
343 lock_res_and_lock(dlmlock);
344 dlmlock->l_ast_data = NULL;
345 cl_object_attr_lock(obj);
347 cl_object_attr_update(env, obj, attr, CAT_KMS);
348 cl_object_attr_unlock(obj);
349 unlock_res_and_lock(dlmlock);
350 cl_object_put(env, obj);
355 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
356 struct ldlm_lock_desc *new, void *data, int flag)
363 case LDLM_CB_BLOCKING: {
364 struct lustre_handle lockh;
366 ldlm_lock2handle(dlmlock, &lockh);
367 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
372 case LDLM_CB_CANCELING: {
377 * This can be called in the context of outer IO, e.g.,
379 * osc_enqueue_base()->...
380 * ->ldlm_prep_elc_req()->...
381 * ->ldlm_cancel_callback()->...
382 * ->osc_ldlm_blocking_ast()
384 * new environment has to be created to not corrupt outer
387 env = cl_env_get(&refcheck);
393 rc = mdc_dlm_blocking_ast0(env, dlmlock, flag);
394 cl_env_put(env, &refcheck);
404 * Updates object attributes from a lock value block (lvb) received together
405 * with the DLM lock reply from the server.
406 * This can be optimized to not update attributes when lock is a result of a
409 * Called under lock and resource spin-locks.
411 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
412 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
414 struct cl_object *obj = osc2cl(osc);
415 struct lov_oinfo *oinfo = osc->oo_oinfo;
416 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
417 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
419 unsigned int setkms = 0;
424 LASSERT(dlmlock != NULL);
425 lvb = &dlmlock->l_ost_lvb;
427 cl_lvb2attr(attr, lvb);
429 cl_object_attr_lock(obj);
430 if (dlmlock != NULL) {
433 check_res_locked(dlmlock->l_resource);
434 size = lvb->lvb_size;
436 if (size >= oinfo->loi_kms) {
438 attr->cat_kms = size;
443 /* The size should not be less than the kms */
444 if (attr->cat_size < oinfo->loi_kms)
445 attr->cat_size = oinfo->loi_kms;
447 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
448 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
449 setkms ? "" : " leaving",
450 setkms ? attr->cat_kms : oinfo->loi_kms,
451 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
453 cl_object_attr_update(env, obj, attr, valid);
454 cl_object_attr_unlock(obj);
458 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
459 struct lustre_handle *lockh)
461 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
462 struct ldlm_lock *dlmlock;
466 dlmlock = ldlm_handle2lock_long(lockh, 0);
467 LASSERT(dlmlock != NULL);
469 /* lock reference taken by ldlm_handle2lock_long() is
470 * owned by osc_lock and released in osc_lock_detach()
472 lu_ref_add(&dlmlock->l_reference, "osc_lock", oscl);
473 oscl->ols_has_ref = 1;
475 LASSERT(oscl->ols_dlmlock == NULL);
476 oscl->ols_dlmlock = dlmlock;
478 /* This may be a matched lock for glimpse request, do not hold
479 * lock reference in that case. */
480 if (!oscl->ols_glimpse) {
481 /* hold a refc for non glimpse lock which will
482 * be released in osc_lock_cancel() */
483 lustre_handle_copy(&oscl->ols_handle, lockh);
484 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
488 /* Lock must have been granted. */
489 lock_res_and_lock(dlmlock);
490 if (dlmlock->l_granted_mode == dlmlock->l_req_mode) {
491 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
493 /* extend the lock extent, otherwise it will have problem when
494 * we decide whether to grant a lockless lock. */
495 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
496 descr->cld_start = cl_index(descr->cld_obj, 0);
497 descr->cld_end = CL_PAGE_EOF;
499 /* no lvb update for matched lock */
500 if (!ldlm_is_lvb_cached(dlmlock)) {
501 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
502 LASSERT(osc == dlmlock->l_ast_data);
503 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
504 ldlm_set_lvb_cached(dlmlock);
507 unlock_res_and_lock(dlmlock);
509 LASSERT(oscl->ols_state != OLS_GRANTED);
510 oscl->ols_state = OLS_GRANTED;
515 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
516 * received from a server, or after osc_enqueue_base() matched a local DLM
519 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
522 struct osc_lock *oscl = cookie;
523 struct cl_lock_slice *slice = &oscl->ols_cl;
529 env = cl_env_percpu_get();
530 /* should never happen, similar to osc_ldlm_blocking_ast(). */
531 LASSERT(!IS_ERR(env));
533 rc = ldlm_error2errno(errcode);
534 if (oscl->ols_state == OLS_ENQUEUED) {
535 oscl->ols_state = OLS_UPCALL_RECEIVED;
536 } else if (oscl->ols_state == OLS_CANCELLED) {
539 CERROR("Impossible state: %d\n", oscl->ols_state);
543 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
545 mdc_lock_granted(env, oscl, lockh);
547 /* Error handling, some errors are tolerable. */
548 if (oscl->ols_locklessable && rc == -EUSERS) {
549 /* This is a tolerable error, turn this lock into
552 osc_object_set_contended(cl2osc(slice->cls_obj));
553 LASSERT(slice->cls_ops != oscl->ols_lockless_ops);
555 /* Change this lock to ldlmlock-less lock. */
556 osc_lock_to_lockless(env, oscl, 1);
557 oscl->ols_state = OLS_GRANTED;
559 } else if (oscl->ols_glimpse && rc == -ENAVAIL) {
560 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
561 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
562 NULL, &oscl->ols_lvb);
563 /* Hide the error. */
567 if (oscl->ols_owner != NULL)
568 cl_sync_io_note(env, oscl->ols_owner, rc);
569 cl_env_percpu_put(env);
574 int mdc_fill_lvb(struct ptlrpc_request *req, struct ost_lvb *lvb)
576 struct mdt_body *body;
578 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
582 lvb->lvb_mtime = body->mbo_mtime;
583 lvb->lvb_atime = body->mbo_atime;
584 lvb->lvb_ctime = body->mbo_ctime;
585 lvb->lvb_blocks = body->mbo_dom_blocks;
586 lvb->lvb_size = body->mbo_dom_size;
591 int mdc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
592 void *cookie, struct lustre_handle *lockh,
593 enum ldlm_mode mode, __u64 *flags, int errcode)
595 struct osc_lock *ols = cookie;
596 struct ldlm_lock *lock;
601 /* The request was created before ldlm_cli_enqueue call. */
602 if (errcode == ELDLM_LOCK_ABORTED) {
603 struct ldlm_reply *rep;
605 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
606 LASSERT(rep != NULL);
608 rep->lock_policy_res2 =
609 ptlrpc_status_ntoh(rep->lock_policy_res2);
610 if (rep->lock_policy_res2)
611 errcode = rep->lock_policy_res2;
613 rc = mdc_fill_lvb(req, &ols->ols_lvb);
614 *flags |= LDLM_FL_LVB_READY;
615 } else if (errcode == ELDLM_OK) {
616 /* Callers have references, should be valid always */
617 lock = ldlm_handle2lock(lockh);
620 rc = mdc_fill_lvb(req, &lock->l_ost_lvb);
622 *flags |= LDLM_FL_LVB_READY;
625 /* Call the update callback. */
626 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
628 /* release the reference taken in ldlm_cli_enqueue() */
629 if (errcode == ELDLM_LOCK_MATCHED)
631 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
632 ldlm_lock_decref(lockh, mode);
637 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
640 struct osc_enqueue_args *aa = args;
641 struct ldlm_lock *lock;
642 struct lustre_handle *lockh = &aa->oa_lockh;
643 enum ldlm_mode mode = aa->oa_mode;
647 LASSERT(!aa->oa_speculative);
649 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
651 lock = ldlm_handle2lock(lockh);
652 LASSERTF(lock != NULL,
653 "lockh %#llx, req %p, aa %p - client evicted?\n",
654 lockh->cookie, req, aa);
656 /* Take an additional reference so that a blocking AST that
657 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
658 * to arrive after an upcall has been executed by
659 * osc_enqueue_fini(). */
660 ldlm_lock_addref(lockh, mode);
662 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
663 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
665 /* Let CP AST to grant the lock first. */
666 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
668 /* Complete obtaining the lock procedure. */
669 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
670 aa->oa_mode, aa->oa_flags, NULL, 0,
672 /* Complete mdc stuff. */
673 rc = mdc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
676 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
678 ldlm_lock_decref(lockh, mode);
683 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
684 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
685 * other synchronous requests, however keeping some locks and trying to obtain
686 * others may take a considerable amount of time in a case of ost failure; and
687 * when other sync requests do not get released lock from a client, the client
688 * is excluded from the cluster -- such scenarious make the life difficult, so
689 * release locks just after they are obtained. */
690 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
691 struct ldlm_res_id *res_id, __u64 *flags,
692 union ldlm_policy_data *policy,
693 struct ost_lvb *lvb, int kms_valid,
694 osc_enqueue_upcall_f upcall, void *cookie,
695 struct ldlm_enqueue_info *einfo, int async)
697 struct obd_device *obd = exp->exp_obd;
698 struct lustre_handle lockh = { 0 };
699 struct ptlrpc_request *req = NULL;
700 struct ldlm_intent *lit;
702 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
703 __u64 match_flags = *flags;
709 mode = einfo->ei_mode;
710 if (einfo->ei_mode == LCK_PR)
714 match_flags |= LDLM_FL_BLOCK_GRANTED;
715 /* DOM locking uses LDLM_FL_KMS_IGNORE to mark locks wich have no valid
716 * LVB information, e.g. canceled locks or locks of just pruned object,
717 * such locks should be skipped.
719 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
720 einfo->ei_type, policy, mode, &lockh);
722 struct ldlm_lock *matched;
724 if (*flags & LDLM_FL_TEST_LOCK)
727 matched = ldlm_handle2lock(&lockh);
729 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
730 ldlm_set_kms_ignore(matched);
732 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
733 *flags |= LDLM_FL_LVB_READY;
735 /* We already have a lock, and it's referenced. */
736 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
738 ldlm_lock_decref(&lockh, mode);
739 LDLM_LOCK_PUT(matched);
742 ldlm_lock_decref(&lockh, mode);
743 LDLM_LOCK_PUT(matched);
746 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
749 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_INTENT);
753 /* For WRITE lock cancel other locks on resource early if any */
754 if (einfo->ei_mode & LCK_PW)
755 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
761 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
763 ptlrpc_request_free(req);
767 /* pack the intent */
768 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
769 lit->opc = glimpse ? IT_GLIMPSE : IT_BRW;
771 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
772 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
773 ptlrpc_request_set_replen(req);
775 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
776 *flags &= ~LDLM_FL_BLOCK_GRANTED;
777 /* All MDC IO locks are intents */
778 *flags |= LDLM_FL_HAS_INTENT;
779 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, NULL,
780 0, LVB_T_NONE, &lockh, async);
783 struct osc_enqueue_args *aa;
785 aa = ptlrpc_req_async_args(aa, req);
787 aa->oa_mode = einfo->ei_mode;
788 aa->oa_type = einfo->ei_type;
789 lustre_handle_copy(&aa->oa_lockh, &lockh);
790 aa->oa_upcall = upcall;
791 aa->oa_cookie = cookie;
792 aa->oa_speculative = false;
793 aa->oa_flags = flags;
796 req->rq_interpret_reply = mdc_enqueue_interpret;
797 ptlrpcd_add_req(req);
799 ptlrpc_req_finished(req);
804 rc = mdc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
806 ptlrpc_req_finished(req);
811 * Implementation of cl_lock_operations::clo_enqueue() method for osc
812 * layer. This initiates ldlm enqueue:
814 * - cancels conflicting locks early (osc_lock_enqueue_wait());
816 * - calls osc_enqueue_base() to do actual enqueue.
818 * osc_enqueue_base() is supplied with an upcall function that is executed
819 * when lock is received either after a local cached ldlm lock is matched, or
820 * when a reply from the server is received.
822 * This function does not wait for the network communication to complete.
824 static int mdc_lock_enqueue(const struct lu_env *env,
825 const struct cl_lock_slice *slice,
826 struct cl_io *unused, struct cl_sync_io *anchor)
828 struct osc_thread_info *info = osc_env_info(env);
829 struct osc_io *oio = osc_env_io(env);
830 struct osc_object *osc = cl2osc(slice->cls_obj);
831 struct osc_lock *oscl = cl2osc_lock(slice);
832 struct cl_lock *lock = slice->cls_lock;
833 struct ldlm_res_id *resname = &info->oti_resname;
834 union ldlm_policy_data *policy = &info->oti_policy;
835 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
836 void *cookie = (void *)oscl;
842 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
843 "lock = %p, ols = %p\n", lock, oscl);
845 if (oscl->ols_state == OLS_GRANTED)
848 /* Lockahead is not supported on MDT yet */
849 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
850 result = -EOPNOTSUPP;
854 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
855 GOTO(enqueue_base, 0);
857 if (oscl->ols_glimpse) {
858 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
860 GOTO(enqueue_base, 0);
863 result = osc_lock_enqueue_wait(env, osc, oscl);
867 /* we can grant lockless lock right after all conflicting locks
869 if (osc_lock_is_lockless(oscl)) {
870 oscl->ols_state = OLS_GRANTED;
871 oio->oi_lockless = 1;
876 oscl->ols_state = OLS_ENQUEUED;
877 if (anchor != NULL) {
878 atomic_inc(&anchor->csi_sync_nr);
879 oscl->ols_owner = anchor;
883 * DLM lock's ast data must be osc_object;
884 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
887 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
888 mdc_lock_build_policy(env, policy);
889 LASSERT(!oscl->ols_speculative);
890 result = mdc_enqueue_send(env, osc_export(osc), resname,
891 &oscl->ols_flags, policy,
892 &oscl->ols_lvb, osc->oo_oinfo->loi_kms_valid,
893 upcall, cookie, &oscl->ols_einfo, async);
895 if (osc_lock_is_lockless(oscl)) {
896 oio->oi_lockless = 1;
898 LASSERT(oscl->ols_state == OLS_GRANTED);
899 LASSERT(oscl->ols_hold);
900 LASSERT(oscl->ols_dlmlock != NULL);
905 oscl->ols_state = OLS_CANCELLED;
906 osc_lock_wake_waiters(env, osc, oscl);
909 cl_sync_io_note(env, anchor, result);
914 static const struct cl_lock_operations mdc_lock_lockless_ops = {
915 .clo_fini = osc_lock_fini,
916 .clo_enqueue = mdc_lock_enqueue,
917 .clo_cancel = mdc_lock_lockless_cancel,
918 .clo_print = osc_lock_print
921 static const struct cl_lock_operations mdc_lock_ops = {
922 .clo_fini = osc_lock_fini,
923 .clo_enqueue = mdc_lock_enqueue,
924 .clo_cancel = osc_lock_cancel,
925 .clo_print = osc_lock_print,
928 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
929 struct cl_lock *lock, const struct cl_io *io)
931 struct osc_lock *ols;
932 __u32 enqflags = lock->cll_descr.cld_enq_flags;
933 __u64 flags = osc_enq2ldlm_flags(enqflags);
937 /* Ignore AGL for Data-on-MDT, stat returns size data */
938 if ((enqflags & CEF_SPECULATIVE) != 0)
941 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
942 if (unlikely(ols == NULL))
945 ols->ols_state = OLS_NEW;
946 spin_lock_init(&ols->ols_lock);
947 INIT_LIST_HEAD(&ols->ols_waiting_list);
948 INIT_LIST_HEAD(&ols->ols_wait_entry);
949 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
950 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
952 ols->ols_flags = flags;
953 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
955 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
956 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
957 ols->ols_glimpse = 1;
959 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
961 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
963 if (!(enqflags & CEF_MUST))
964 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
965 if (ols->ols_locklessable && !(enqflags & CEF_DISCARD_DATA))
966 ols->ols_flags |= LDLM_FL_DENY_ON_CONTENTION;
968 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
969 osc_lock_set_writer(env, io, obj, ols);
971 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
972 lock, ols, ols->ols_flags);
979 * An implementation of cl_io_operations specific methods for MDC layer.
982 static int mdc_async_upcall(void *a, int rc)
984 struct osc_async_cbargs *args = a;
987 complete(&args->opc_sync);
991 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
992 pgoff_t index, struct lustre_handle *lh)
994 struct ldlm_lock *lock;
996 /* find DOM lock protecting object */
997 lock = mdc_dlmlock_at_pgoff(env, osc, index,
998 OSC_DAP_FL_TEST_LOCK |
999 OSC_DAP_FL_CANCELING);
1001 struct ldlm_resource *res;
1002 struct ldlm_res_id *resname;
1004 resname = &osc_env_info(env)->oti_resname;
1005 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1006 res = ldlm_resource_get(osc_export(osc)->exp_obd->obd_namespace,
1007 NULL, resname, LDLM_IBITS, 0);
1008 ldlm_resource_dump(D_ERROR, res);
1009 libcfs_debug_dumpstack(NULL);
1012 *lh = lock->l_remote_handle;
1013 LDLM_LOCK_PUT(lock);
1018 static int mdc_io_setattr_start(const struct lu_env *env,
1019 const struct cl_io_slice *slice)
1021 struct cl_io *io = slice->cis_io;
1022 struct osc_io *oio = cl2osc_io(env, slice);
1023 struct cl_object *obj = slice->cis_obj;
1024 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1025 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1026 struct obdo *oa = &oio->oi_oa;
1027 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1028 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1029 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1030 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1033 /* silently ignore non-truncate setattr for Data-on-MDT object */
1034 if (cl_io_is_trunc(io)) {
1035 /* truncate cache dirty pages first */
1036 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1042 if (oio->oi_lockless == 0) {
1043 cl_object_attr_lock(obj);
1044 rc = cl_object_attr_get(env, obj, attr);
1046 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1047 unsigned int cl_valid = 0;
1049 if (ia_avalid & ATTR_SIZE) {
1050 attr->cat_size = size;
1051 attr->cat_kms = size;
1052 cl_valid = (CAT_SIZE | CAT_KMS);
1054 if (ia_avalid & ATTR_MTIME_SET) {
1055 attr->cat_mtime = lvb->lvb_mtime;
1056 cl_valid |= CAT_MTIME;
1058 if (ia_avalid & ATTR_ATIME_SET) {
1059 attr->cat_atime = lvb->lvb_atime;
1060 cl_valid |= CAT_ATIME;
1062 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1063 attr->cat_ctime = lvb->lvb_ctime;
1064 cl_valid |= CAT_CTIME;
1066 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1068 cl_object_attr_unlock(obj);
1073 if (!(ia_avalid & ATTR_SIZE))
1076 memset(oa, 0, sizeof(*oa));
1077 oa->o_oi = loi->loi_oi;
1078 oa->o_mtime = attr->cat_mtime;
1079 oa->o_atime = attr->cat_atime;
1080 oa->o_ctime = attr->cat_ctime;
1083 oa->o_blocks = OBD_OBJECT_EOF;
1084 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1085 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1087 if (oio->oi_lockless) {
1088 oa->o_flags = OBD_FL_SRVLOCK;
1089 oa->o_valid |= OBD_MD_FLFLAGS;
1091 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1094 oa->o_valid |= OBD_MD_FLHANDLE;
1097 init_completion(&cbargs->opc_sync);
1099 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1100 mdc_async_upcall, cbargs);
1101 cbargs->opc_rpc_sent = rc == 0;
1105 static int mdc_io_read_ahead(const struct lu_env *env,
1106 const struct cl_io_slice *ios,
1107 pgoff_t start, struct cl_read_ahead *ra)
1109 struct osc_object *osc = cl2osc(ios->cis_obj);
1110 struct ldlm_lock *dlmlock;
1114 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1115 if (dlmlock == NULL)
1118 if (dlmlock->l_req_mode != LCK_PR) {
1119 struct lustre_handle lockh;
1121 ldlm_lock2handle(dlmlock, &lockh);
1122 ldlm_lock_addref(&lockh, LCK_PR);
1123 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1126 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1127 ra->cra_end_idx = CL_PAGE_EOF;
1128 ra->cra_release = osc_read_ahead_release;
1129 ra->cra_cbdata = dlmlock;
1134 int mdc_io_fsync_start(const struct lu_env *env,
1135 const struct cl_io_slice *slice)
1137 struct cl_io *io = slice->cis_io;
1138 struct cl_fsync_io *fio = &io->u.ci_fsync;
1139 struct cl_object *obj = slice->cis_obj;
1140 struct osc_object *osc = cl2osc(obj);
1145 /* a MDC lock always covers whole object, do sync for whole
1146 * possible range despite of supplied start/end values.
1148 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1149 fio->fi_mode == CL_FSYNC_DISCARD);
1151 fio->fi_nr_written += result;
1154 if (fio->fi_mode == CL_FSYNC_ALL) {
1157 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1160 /* Use OSC sync code because it is asynchronous.
1161 * It is to be added into MDC and avoid the using of
1162 * OST_SYNC at both MDC and MDT.
1164 rc = osc_fsync_ost(env, osc, fio);
1172 struct mdc_data_version_args {
1173 struct osc_io *dva_oio;
1177 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1180 struct mdc_data_version_args *dva = args;
1181 struct osc_io *oio = dva->dva_oio;
1182 const struct mdt_body *body;
1188 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1190 GOTO(out, rc = -EPROTO);
1192 /* Prepare OBDO from mdt_body for CLIO */
1193 oio->oi_oa.o_valid = body->mbo_valid;
1194 oio->oi_oa.o_flags = body->mbo_flags;
1195 oio->oi_oa.o_data_version = body->mbo_version;
1196 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1199 oio->oi_cbarg.opc_rc = rc;
1200 complete(&oio->oi_cbarg.opc_sync);
1204 static int mdc_io_data_version_start(const struct lu_env *env,
1205 const struct cl_io_slice *slice)
1207 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1208 struct osc_io *oio = cl2osc_io(env, slice);
1209 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1210 struct osc_object *obj = cl2osc(slice->cis_obj);
1211 struct obd_export *exp = osc_export(obj);
1212 struct ptlrpc_request *req;
1213 struct mdt_body *body;
1214 struct mdc_data_version_args *dva;
1219 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1220 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1221 oio->oi_oa.o_valid = OBD_MD_FLID;
1223 init_completion(&cbargs->opc_sync);
1225 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1229 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1231 ptlrpc_request_free(req);
1235 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1236 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1237 body->mbo_valid = OBD_MD_FLID;
1238 /* Indicate that data version is needed */
1239 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1240 body->mbo_flags = 0;
1242 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1243 body->mbo_valid |= OBD_MD_FLFLAGS;
1244 body->mbo_flags |= OBD_FL_SRVLOCK;
1245 if (dv->dv_flags & LL_DV_WR_FLUSH)
1246 body->mbo_flags |= OBD_FL_FLUSH;
1249 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1250 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1251 ptlrpc_request_set_replen(req);
1253 req->rq_interpret_reply = mdc_data_version_interpret;
1254 dva = ptlrpc_req_async_args(dva, req);
1257 ptlrpcd_add_req(req);
1262 static void mdc_io_data_version_end(const struct lu_env *env,
1263 const struct cl_io_slice *slice)
1265 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1266 struct osc_io *oio = cl2osc_io(env, slice);
1267 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1270 wait_for_completion(&cbargs->opc_sync);
1272 if (cbargs->opc_rc != 0) {
1273 slice->cis_io->ci_result = cbargs->opc_rc;
1275 slice->cis_io->ci_result = 0;
1276 if (!(oio->oi_oa.o_valid &
1277 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1278 slice->cis_io->ci_result = -ENOTSUPP;
1280 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1281 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1282 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1283 dv->dv_data_version = oio->oi_oa.o_data_version;
1289 static struct cl_io_operations mdc_io_ops = {
1292 .cio_iter_init = osc_io_rw_iter_init,
1293 .cio_iter_fini = osc_io_rw_iter_fini,
1294 .cio_start = osc_io_read_start,
1297 .cio_iter_init = osc_io_rw_iter_init,
1298 .cio_iter_fini = osc_io_rw_iter_fini,
1299 .cio_start = osc_io_write_start,
1300 .cio_end = osc_io_end,
1303 .cio_iter_init = osc_io_iter_init,
1304 .cio_iter_fini = osc_io_iter_fini,
1305 .cio_start = mdc_io_setattr_start,
1306 .cio_end = osc_io_setattr_end,
1308 [CIT_DATA_VERSION] = {
1309 .cio_start = mdc_io_data_version_start,
1310 .cio_end = mdc_io_data_version_end,
1313 .cio_iter_init = osc_io_iter_init,
1314 .cio_iter_fini = osc_io_iter_fini,
1315 .cio_start = osc_io_fault_start,
1316 .cio_end = osc_io_end,
1319 .cio_start = mdc_io_fsync_start,
1320 .cio_end = osc_io_fsync_end,
1323 .cio_read_ahead = mdc_io_read_ahead,
1324 .cio_submit = osc_io_submit,
1325 .cio_commit_async = osc_io_commit_async,
1328 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1331 struct osc_io *oio = osc_env_io(env);
1333 CL_IO_SLICE_CLEAN(oio, oi_cl);
1334 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1338 static void mdc_build_res_name(struct osc_object *osc,
1339 struct ldlm_res_id *resname)
1341 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1345 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1346 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1349 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1350 struct cl_req_attr *attr)
1352 u64 flags = attr->cra_flags;
1354 /* Copy object FID to cl_attr */
1355 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1357 if (flags & OBD_MD_FLGROUP)
1358 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1360 if (flags & OBD_MD_FLID)
1361 attr->cra_oa->o_valid |= OBD_MD_FLID;
1363 if (flags & OBD_MD_FLHANDLE) {
1364 struct osc_page *opg;
1366 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1367 if (!opg->ops_srvlock) {
1370 rc = mdc_get_lock_handle(env, cl2osc(obj),
1372 &attr->cra_oa->o_handle);
1374 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1375 "uncovered page!\n");
1378 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1384 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1385 struct cl_attr *attr)
1387 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1389 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1390 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1392 return osc_attr_get(env, obj, attr);
1395 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1397 struct osc_object *osc = (struct osc_object *)data;
1398 struct ost_lvb *lvb = &lock->l_ost_lvb;
1399 struct lov_oinfo *oinfo;
1402 if (lock->l_ast_data == data) {
1403 lock->l_ast_data = NULL;
1405 LASSERT(osc != NULL);
1406 LASSERT(osc->oo_oinfo != NULL);
1407 LASSERT(lvb != NULL);
1409 /* Updates lvb in lock by the cached oinfo */
1410 oinfo = osc->oo_oinfo;
1412 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1413 "%llu %llu %llu by oinfo size %llu blocks %llu "
1414 "[cma]time %llu %llu %llu", lvb->lvb_size,
1415 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1416 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1417 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1418 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1419 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1421 cl_object_attr_lock(&osc->oo_cl);
1422 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1423 cl_object_attr_unlock(&osc->oo_cl);
1424 ldlm_clear_lvb_cached(lock);
1426 RETURN(LDLM_ITER_CONTINUE);
1429 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1431 struct osc_object *osc = cl2osc(obj);
1432 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1434 /* DLM locks don't hold a reference of osc_object so we have to
1435 * clear it before the object is being destroyed. */
1436 osc_build_res_name(osc, resname);
1437 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1438 mdc_object_ast_clear, osc);
1442 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1443 struct ldlm_lock *lock)
1445 /* if lock cancel is initiated from llite then it is combined
1446 * lock with DOM bit and it may have no l_ast_data initialized yet,
1447 * so init it here with given osc_object.
1449 mdc_set_dom_lock_data(lock, cl2osc(obj));
1450 RETURN(mdc_dlm_blocking_ast0(env, lock, LDLM_CB_CANCELING));
1453 static const struct cl_object_operations mdc_ops = {
1454 .coo_page_init = osc_page_init,
1455 .coo_lock_init = mdc_lock_init,
1456 .coo_io_init = mdc_io_init,
1457 .coo_attr_get = mdc_attr_get,
1458 .coo_attr_update = osc_attr_update,
1459 .coo_glimpse = osc_object_glimpse,
1460 .coo_req_attr_set = mdc_req_attr_set,
1461 .coo_prune = mdc_object_prune,
1462 .coo_object_flush = mdc_object_flush
1465 static const struct osc_object_operations mdc_object_ops = {
1466 .oto_build_res_name = mdc_build_res_name,
1467 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1470 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1471 const struct lu_object_conf *conf)
1473 struct osc_object *osc = lu2osc(obj);
1475 if (osc->oo_initialized)
1478 osc->oo_initialized = true;
1480 return osc_object_init(env, obj, conf);
1483 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1485 osc_object_free(env, obj);
1488 static const struct lu_object_operations mdc_lu_obj_ops = {
1489 .loo_object_init = mdc_object_init,
1490 .loo_object_delete = NULL,
1491 .loo_object_release = NULL,
1492 .loo_object_free = mdc_object_free,
1493 .loo_object_print = osc_object_print,
1494 .loo_object_invariant = NULL
1497 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1498 const struct lu_object_header *unused,
1499 struct lu_device *dev)
1501 struct osc_object *osc;
1502 struct lu_object *obj;
1504 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1507 lu_object_init(obj, NULL, dev);
1508 osc->oo_cl.co_ops = &mdc_ops;
1509 obj->lo_ops = &mdc_lu_obj_ops;
1510 osc->oo_obj_ops = &mdc_object_ops;
1511 osc->oo_initialized = false;
1518 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1519 struct lustre_cfg *cfg)
1521 size_t count = class_modify_config(cfg, PARAM_MDC,
1522 &d->ld_obd->obd_kset.kobj);
1523 return count > 0 ? 0 : count;
1526 const struct lu_device_operations mdc_lu_ops = {
1527 .ldo_object_alloc = mdc_object_alloc,
1528 .ldo_process_config = mdc_process_config,
1529 .ldo_recovery_complete = NULL,
1532 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1533 struct lu_device_type *t,
1534 struct lustre_cfg *cfg)
1536 struct lu_device *d;
1537 struct osc_device *od;
1538 struct obd_device *obd;
1543 RETURN(ERR_PTR(-ENOMEM));
1545 cl_device_init(&od->od_cl, t);
1547 d->ld_ops = &mdc_lu_ops;
1550 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1552 RETURN(ERR_PTR(-ENODEV));
1554 rc = mdc_setup(obd, cfg);
1556 osc_device_free(env, d);
1557 RETURN(ERR_PTR(rc));
1559 od->od_exp = obd->obd_self_export;
1563 static const struct lu_device_type_operations mdc_device_type_ops = {
1564 .ldto_device_alloc = mdc_device_alloc,
1565 .ldto_device_free = osc_device_free,
1566 .ldto_device_init = osc_device_init,
1567 .ldto_device_fini = osc_device_fini
1570 struct lu_device_type mdc_device_type = {
1571 .ldt_tags = LU_DEVICE_CL,
1572 .ldt_name = LUSTRE_MDC_NAME,
1573 .ldt_ops = &mdc_device_type_ops,
1574 .ldt_ctx_tags = LCT_CL_THREAD